AMD: Polaris 10 targets mainstream desktop & high-end notebook

[raghu78]

AMD has updated their projections to 2x times performance per watt and no longer 2.5x. And this is over their mainstream products and not stuff like nano or fury. So if AMD makes the same improvements as Nvidia since they both claim 2x increase in performance per watt, they are going to be in the same position as now.

AMD CEO Lisa Su has stated 2x per/watt gains earlier too. But Raja Koduri at Capsaicin stated 2.5x perf/watt gains for Polaris over 28nm GPUs. Given that Raja heads RTG I am going to go with his statements. Lisa Su is a CEO and is just reading out prepared statements. You can make of it what you want to.

 

[renderstate]

There are some situation where Polaris is incredibly fast. Faster than anything in the market. The secret is probably that special culling mechanism in the hardware, which helps the GPU to effectively cull those false positive primitives that aren't visible in the screen. Today's hardwares can't do this.

Saying that today's hardware doesn't discard non visible primitives is non sense (of course it does). Perhaps what you wanted to say is that some HW out there is not very fast at culling primitives that do not contribute to the final image (which is why some developers write their own CPU based or compute based triangle culling pipelines).

Excluding triangles occluded by other geometry 3D APIs require support scissoring, clipping and back face culling. Moreover it is trivial to cull zero area/degenerate triangles. The hardest case is long and thin triangles that happen to not cover any screen space sample. Some HW does take care of some of these too before they reach the rasterizer.

I have no idea what Polaris primitive discard unit does but from the sound of it they might have addressed some sub-optimal cases and overall increased their culling rate to match other architectures. Something smart they could do is compute the extent of a geometry packet and cull it altogether if it is outside the frustum or not facing the camera.

Single wavefront perfomance is also incredibly good. 10-100 times faster than anything in the market. This is good for VR.

Assuming that single wavefront performance is as fast as you say (proof?), why would it be so relevant for VR?

 

[JDG1980]

Ya, so you weren't reading my post at all were you? Read again what I said. 7850 is nowhere close to a 75W card either.

As indicated in the link I posted, TPU tested the 7850 as having an average gaming power draw of 87W. Obviously that isn't the maximum possible power draw (that's a bit over 100W), but the demo run by AMD also isn't maximum power usage; it's a game being run with unknown-to-us settings. If it's locked to 60 FPS, then the power usage may well be quite a bit below its maximum TDP.

And the truth is that we're all making educated guesses on perf/watt comparisons. You're assuming a comparison to the existing Hawaii SKUs, which is pretty much a worst-case scenario because these are being pushed balls-to-the-wall and are thus very inefficient. (Hawaii provides about 3/4-7/8 the performance at 1/2 the power if you lower the TDP limit.) I am tentatively assuming AMD is comparing products to those of similar positioning in their 28nm stack: Pitcairn to Polaris 10, Cape Verde to Polaris 11, and so forth.

That's now what I am saying. I am saying Polaris 10 may have a chance to match or beat 980's performance but not a 75W Polaris 10. I am not saying anything about a 150W Polaris 10 not being able to match a 980.

I agree a 75W Polaris won't beat a GTX 980. But I don't think it will be as far off as you suspect. I also think that we will see bigger gains in minimum FPS than average FPS, due to the specific improvements mentioned in Polaris and the fact that AMD appeared to emphasize this in the Hitman demo.

 

[kapulek]

AMD CEO Lisa Su has stated 2x per/watt gains earlier too. But Raja Koduri at Capsaicin stated 2.5x perf/watt gains for Polaris over 28nm GPUs. Given that Raja heads RTG I am going to go with his statements. Lisa Su is a CEO and is just reading out prepared statements. You can make of it what you want to.

Raja Koduri never stated 2.5x perf/W officially.

 

[Tuna-Fish]

AMD CEO Lisa Su has stated 2x per/watt gains earlier too. But Raja Koduri at Capsaicin stated 2.5x perf/watt gains for Polaris over 28nm GPUs. Given that Raja heads RTG I am going to go with his statements. Lisa Su is a CEO and is just reading out prepared statements. You can make of it what you want to.

The real difference is that if you get caught lying to customers, you might hurt some feelings, possibly lose some sales or in the very worst case the company has to pay some settlement, and maybe the company fires you to save face. However, when c-level employees get caught lying to investors, they go to prison. They can be held personally liable for damages. In this situation, they don't just get fired, they might lose every penny they own. The non-committal corporate speak that is used in these events evolved precisely because lying can get your ass thrown in jail. I would always take the prepared statement read by a marketing drone in an investor relations event about a 1000x more seriously than whatever an engineer says about his creation. Because you can be damn sure that the prepared statement has gone through the engineers and been checked.

The most charitable explanation for the 2.5x/2x divide is that the origin of the comparison differs -- 2x is from Fiji and 2.5x is from Tonga or something. However, at this point I would personally prefer to use just 2x.

 

[Abwx]

Raja Koduri never stated 2.5x perf/W officially.

He didnt say so..?..

Yet he had this pic in front of his eyes during his last conf, i guess that we ll have to wait for the next such event...

The most charitable explanation for the 2.5x/2x divide is that the origin of the comparison differs -- 2x is from Fiji and 2.5x is from Tonga or something. However, at this point I would personally prefer to use just 2x.

I would say 3.5x for a die shrink and the same throughput (read frequency and SP count), but that s just my opinion...

 

[tential]

The real difference is that if you get caught lying to customers, you might hurt some feelings, possibly lose some sales or in the very worst case the company has to pay some settlement, and maybe the company fires you to save face. However, when c-level employees get caught lying to investors, they go to prison. They can be held personally liable for damages. In this situation, they don't just get fired, they might lose every penny they own. The non-committal corporate speak that is used in these events evolved precisely because lying can get your ass thrown in jail. I would always take the prepared statement read by a marketing drone in an investor relations event about a 1000x more seriously than whatever an engineer says about his creation. Because you can be damn sure that the prepared statement has gone through the engineers and been checked.

The most charitable explanation for the 2.5x/2x divide is that the origin of the comparison differs -- 2x is from Fiji and 2.5x is from Tonga or something. However, at this point I would personally prefer to use just 2x.

Just like those prepared statements that Nvidia gave about the GTX 970 right?

Anyone can be wrong, get over it.

 

[IllogicalGlory]

Raja Koduri never stated 2.5x perf/W officially.

It's on the road map they showed at Capsaicin, and indeed Raja Koduri does say it himself at the same time.

14:08 "We just announced Polaris at 2.5 times performance per watt." He repeats it before the Hitman demo as well.

 

[Tuna-Fish]

Just like those prepared statements that Nvidia gave about the GTX 970 right?

During which investor relations event did nVidia make statements about the 970?

 

[flopper]

AMD CEO Lisa Su has stated 2x per/watt gains earlier too. But Raja Koduri at Capsaicin stated 2.5x perf/watt gains for Polaris over 28nm GPUs. Given that Raja heads RTG I am going to go with his statements. Lisa Su is a CEO and is just reading out prepared statements. You can make of it what you want to.

she is also smart and an engineer.

 

[3DVagabond]

she is also smart and an engineer.

Yeah. She's not your typical marketing person at the top (who can also be an engineer). She hasn't shown the propensity to showboat.

That said I would assume any prediction is "best case" and not average.

 

[flopper]

Yeah. She's not your typical marketing person at the top (who can also be an engineer). She hasn't shown the propensity to showboat.

That said I would assume any prediction is "best case" and not average.

Obviously if it was average, be prepared to be blown away
might happen anyway which is why this launch of Polaris might be a real postmark in history

 

[Silverforce11]

Don't get all twisty over the 2.5x because we don't know the comparison.

2.5x vs Hawaii is okay. But 2.5x vs Nano? That would be nuts.

At this point, the only valid evidence is this: AMD have said several times, they want to bring higher performance than the minimum required for VR (970 & 290) at a lower price.

So we can be confident it won't be slower than a 970 and 290 and it won't be priced higher than the 970 which is $330.

 

[xLegenday]

Have you guys seen this video?
https://www.youtube.com/watch?v=aSYBO1BrB1I

Pretty interesting, talking how AMD plans to use their console advantage in their future Gen GPUs.

 

[Silverforce11]

Have you guys seen this video?
https://www.youtube.com/watch?v=aSYBO1BrB1I

Pretty interesting, talking how AMD plans to use their console advantage in their future Gen GPUs.

Yes, I posted it here:

http://forums.anandtech.com/showpost.php?p=38183411&postcount=29

It's an interesting video from @Adored but unlike him, I see some real easy ways for NVIDIA to negate AMD's console advantage.

 

[flopper]

Yes, I posted it here:

http://forums.anandtech.com/showpost.php?p=38183411&postcount=29

It's an interesting video from @Adored but unlike him, I see some real easy ways for NVIDIA to negate AMD's console advantage.

Bribes, assassinations?

 

[parkerface]

Bribes, assassinations?

Who needs those when you can send a few product examples to the right places and sprinkle some hashtag twitter glitter on top?

 

[Glo.]

Easiest way for Nvidia to win? Make great hardware. Simple as it can fu**** be.

Best hardware has to win.

 

[krumme]

Sold my 970 and is waiting for polaris 130w. Dont game atm. The old 7970 that i gave to one of the kids was approaching the 970 anyway and faster than anticipated. Absolutely want ps 4.5 hw this round. Will take Vega when it comes and kid can get the 130w. I dont game, he does all the time, so that way we save a lot of electricity. Lol.

Blower might be good with only 130w?

 

[tential]

Yes, I posted it here:

http://forums.anandtech.com/showpost.php?p=38183411&postcount=29

It's an interesting video from @Adored but unlike him, I see some real easy ways for NVIDIA to negate AMD's console advantage.

Nvidia just has a better marketing division. I mean even when I see Nvidia ads on twitch vs amd the Nvidia ones are far more polished a d just better adverts.

If I purchased gpus like I used to off adverts, I'd still buy Nvidia.

 

[EightySix Four]

Best hardware has to win.

Lol

 

[Glo.]

Lol

In other words, for those who do can not understand words: The hardware that brings us the highest performance with lowest possible power consumption always has to win. Regardless of brand.

 

[Arachnotronic]

Just like those prepared statements that Nvidia gave about the GTX 970 right?

Anyone can be wrong, get over it.

No, Tuna-Fish is absolutely right. IF you are a C-level executive and you knowingly lie/mislead investors, that really can mean time in the slammer.

This is why these statements are always peppered with phrases such as:

- "We believe"
- "We expect"
- "We intend"

By using these, execs can wind up being absolutely dead wrong but face no liability because of the "safe harbor" statement provided before these statements are made.

 

[Erenhardt]

In other words, for those who do can not understand words: The hardware that brings us the highest performance with lowest possible power consumption always has to win. Regardless of brand.

I recon you speak for yourself and adress yourself in plural for whatever reason.

Seems you missed a couple of last pages and multiple recent treads. AMD that covered with hd5000 and hd6000, and it was only good for up to 50% marketshare. Meanwhile, it takes the other one whatever metric to get 70%+ of consumers.

It may be good for amd if they release new hardware aiming different segments than their competitor.

 

[Kenmitch]

In other words, for those who do can not understand words: The hardware that brings us the highest performance with lowest possible power consumption always has to win. Regardless of brand.

As a general statement it does seem somewhat misleading. Are you talking relative to price points? Performance classes? DX performance? Mining? Past performance trends? Many factors to consider when choosing a gpu in the end when you think about it.